1. Field of the Invention
The present invention relates to an image processing technique, and more particularly to a technique for detecting an object in an image.
2. Description of the Background Art
Many techniques for detecting an object in an image by pattern matching are proposed. Japanese Patent Application Laid Open Gazette No. 2007-26400 (Patent Document 1), for example, discloses a technique in which pattern matching is performed by using an image region (image to be checked) included in a captured image and an image pattern (template image (reference image)) which is stored in advance, to thereby extract a recognition object from the captured image on the basis of the result of pattern matching.
In the case where a recognition object (predetermined object) is extracted from an image by performing pattern matching, like in the technique of Patent Document 1, usually, it is determined whether or not it is an image (image region) including the recognition object by comparing a matching value calculated through the pattern matching with a predetermined threshold value.
When the pattern matching is performed on the image (image region) including the recognition object is uneven, such as images with different contrasts or images with different distributions of image feature amounts (for example, luminance distributions), the matching value calculated thereby significantly varies. This makes it difficult to specify (extract) a region including the recognition object by using the matching value. In such a case, in order to appropriately perform the pattern matching, it is necessary to prepare a plurality of template images (reference images) for each image characteristic, and further prepare a plurality of threshold values to be compared with the matching values for the plurality of template images.
Meanwhile, there is a technique for specifying (extracting) a bilaterally symmetrical object in an image by using the pattern matching. In such a technique, an image region to be processed is divided by the symmetry axis into one divided image region to be defined as a determination image region and the other divided image region to be defined as a reference image region (template image region), and the pattern matching is performed on the determination image region and the reference image region. Then, by comparing a matching value obtained by the pattern matching with a predetermined threshold value, it is determined whether or not the image region to be processed is a region including a bilaterally symmetrical object.
The above technique does not require to prepare a plurality of reference images. However, if the above technique is applied to images in which image regions to be processed are uneven (for example, image regions with different contrasts or with different distributions of image feature amounts (e.g., luminance distributions)), the matching value obtained through the pattern matching significantly varies. Therefore, it is necessary to perform a complicated thresholding (for example, processing using a plurality of threshold values with grouping in accordance with cases for each image characteristic, or the like).
To solve the above problems, it is an object of the present invention to provide an object detection apparatus, a program, and an integrated circuit each of which is capable of appropriately detecting an axially symmetric object in an image, whatever image is to be processed, without performing any complicated thresholding.